optimize-gpt-6-astra-prompts

Optimize GPT-6 Astra Prompts

Turn rough ideas and existing prompt stacks into copy-ready GPT-6 Astra prompts. Preserve the user's intent and hard constraints, ask for genuinely blocking information, and apply Astra-specific mitigations only where the task or observed behavior requires them.

Use the official basis

Read references/official-guidance.md when model-specific rationale, API-setting guidance, migration notes, or the longer checklist is needed. Treat its URLs as canonical sources and its prose as a dated summary. Fetch the live OpenAI pages before making claims about the current model alias, parameters, availability, or other facts that may change.

For cleaning repository Skills, AGENTS.md, or Sol-era coding-agent scaffolding (wrong skill loads, full-repo pre-reads, over-testing, early stops), use $audit-gpt-6-astra-skills instead of stuffing harness cleanup into a prompt rewrite. That skill follows Rethinking skills and prompts for GPT-6 Astra.

Follow the workflow

1. Capture the prompt contract

Identify the following from the user's message, pasted prompt, or referenced file:

• intended user-visible outcome;
• inputs and context the model will receive;
• hard constraints and facts to preserve;
• required output shape, audience, and language;
• evidence, tool use, or validation needed for success;
• side effects the model may take and actions that require approval;
• autonomy level: when to ask, when to persist, and when approval is the final step;
• fallback or stopping behavior when required information is unavailable.

Treat these as diagnostic dimensions, not mandatory sections. Keep a simple rewriting prompt simple.

If the user supplies a layered prompt stack, preserve the separation between system, developer, user, and tool-description content. Do not silently move instructions between trust levels.

2. Apply the clarity gate